L-Indistinguishability on Eigenvarieties

arXiv: 1508.06187v2 [math.NT] 23 Mar 2016

L-INDISTINGUISHABILITY ON EIGENVARIETIES

JUDITH LUDWIG

Abstract. In this article, we construct examples of L-indistinguishable overcon-
vergent eigenforms for an inner form of SL 2 .

1. Introduction

The purpose of this article is to show the existence of points on eigenvarieties for an
inner form of SL2, whose associated systems of Hecke eigenvalues come from classical
automorphic representations of an inner form of GL2, but which are not classical
themselves, in the sense that there are no classical forms for these systems of Hecke
eigenvalues in the corresponding spaces of overconvergent forms. By construction
there are overconvergent forms giving rise to these points. For each such point we also
construct a twin on a different eigenvariety which is classical. The overconvergent
non-classical form / on the first eigenvariety and the classical form g on the second
eigenvariety are L-indistinguishable in the sense that they give rise to the same
Galois representation. Although there is no definition of a global p-adic L-packet,
our results suggest that, for any future definition, / and g should lie in the same
L-packet.

We describe our results in more detail: Let B/Qhe a definite quaternion algebra
and denote hy Sb the set of primes where B ramifies. Let G be the algebraic group
over Q defined by the units B* and G the subgroup of elements of reduced norm
one. Fix a prime p ^ Sb and a finite extension E/Qp.

For S a finite set of places which includes p and Sb, we have a Hecke algebra
Tis ■= 'Hur,s®E^p for G, which is the product of the spherical Hecke algebras at all
places not in S and an Atkin-Lehner algebra at p, and an analogue for G. Any

idempotent e = ®ei G C'^(G(Ap,Q), such that e; = lGL 2 (Zi) fo'^ ^ ^ S, gives
rise to an eigenvariety L>(e) of idempotent type e, whose underlying set of points
embeds

^(?)(Qp) Hom(?^s,Qp) X W(Qp),

where W = Homcts((Zp^, Gm) denotes the usual weight space. If e G C^(G(Ap, Q)
is an idempotent with the same set S of bad places, we have an eigenvariety L’(e)
of idempotent type e for G, whose underlying set of points embeds

Hom(iI^s,Qp) X >V(Qp),

Date: March 24, 2016.

1

2

JUDITH LUDWIG

for the corresponding weight space W = Hornes(Z*, Gm)- There are natural maps

ns^ils, W.

Definition (Definition 15. ip . A point z on an eigenvariety of idempotent type is
called classical, if there is a classical automorphic eigenform in the corresponding
space of overconvergent forms, whose system of Hecke eigenvalues is that defined
by z.

Let tt{ 9 ) be an algebraic automorphic representation of G{A) associated to a
Grofiencharacter 9 of an imaginary quadratic held L. Assume that p splits in L.
Then such a representation gives rise to two points on 'Dife) for a suitable idempotent
hGCr(G'(Ap,E), one which is ordinary and one which is of critical slope.

Let X be the point of critical slope and consider its image in Hom(^5, Qp) xW{Qp)
under the composite of the maps

P(g)(Qp)-> Hom(Hs,Qp) X >V(Qp)

Hom(i^S,Qp) X W(Qp),

which we denote by (f)-

Our main theorem is the following.

Theorem (Theorem 14 . 3 p . There exist automorphic representations vr( 0 ) of G{A)
as above together with idempotents e G G^{G{A^j,),Q) and ei,e2 G G^{G{A^j),Q),
such that, using the above notation, the image (fifx) of the critical slope refinement x
of 7 r{ 9 ) lifts to a non-classical point on the eigenvariety T){ei) and to a classical
point on V{e2)-

The proof of the theorem uses a p-adic version of a Labesse-Langlands trans¬
fer proved by the author in | 10 j . The representations 7r(0) and the idempotents ei
and 62 are constructed in such a way, that each Ci sees exactly one member of the
L-packet n(7r(0)), i.e., for i = 1,2 there exists a unique element vii G n(7r(0)) such
that ei( 7 ri)j / 0 . Moreover 771(711), the multiplicity of tti in the automorphic spec¬
trum of G, is zero and 772 is automorphic. In particular, this implies that (p{x) lifts
to P(e2). In order to show that it also lifts to P(ei), the crucial point is that in a
neighbourhood of x we can hnd many points associated to automorphic representa¬
tions that do not come from a Grofiencharacter. These automorphic representations
of G{A) give rise to stable L-packets of G and therefore their images under (f all lift
to T’(ei).

By construction, points on eigenvarieties give rise to systems of Hecke eigenvalues
occurring in spaces of overconvergent automorphic forms. So in particular the the¬
orem shows the existence of an overconvergent eigenform / of tame level ei, whose
system of Hecke eigenvalues V’/ comes from a classical automorphic representation
of G{A). The multiplicity formulae of Labesse and Langlands rule out that there is

L-INDISTINGUISHABILITY ON EIGENVARIETIES

3

a classical eigenform for of the same tame level. On the other hand the point
on T>(e2) comes from a classical form, say g. The associated Galois representations
Pf and pg agree as the systems of ^s'-eigenvalues for / and g are the same. In this
sense the two forms are L-indistinguishable.

Notation. Fix embeddings too : Q C and ip : Q Qp as well as a finite
extension E/Qp with ring of integers Oe and an embedding E C Qp. For a number
held K we denote by Gk ■= Gal{K/K) the absolute Galois group of an algebraic
closure of K. For a hnite set S of places of K, let Gk,s be the Galois group of a
maximal extension of K that is unramihed outside S, and for v ^ S, let Frob^ G Gk,s
denote a representative of the geometric Frobenius at v.

We will frequently choose idempotents e G C'“(i?(Ap,Q) where H is equal to

either G or to G. We will always assume that E is chosen big enough so that
ip o e takes values in E. We ease notation by dropping the embeddings from the
notation when it is obvious, e.g., for a complex representation tTj of FI(Ap we write
e • TTj instead of (ioo o e) • tTj. Furthermore we assume that the idempotents we
consider are given as a tensor product of local idempotents ei G G^{H{Qi),Q),
where ei = for almost all 1 . We denote by S{e) the minimal hnite set of hnite

primes containing Sb and p, such that e; = 'i-Hi’Zi) for I ^ S{e).

For a rigid analytic space X dehned over E, any point is dehned over a hnite
extension of E. We write X(Qp) ;= |J^, X{E').

Acknowledgements. The author would like to thank Gaetan Ghenevier, Eugen
Hellmann, James Newton, Peter Scholze and Benjamin Schraen for helpful conver¬
sations and Kevin Buzzard for his comments on an earlier draft of this paper. The
author would also like to thank the anonymous referee for various helpful comments
and suggestions. The author was supported by the SFB/TR 45 of the DFG.

2. Eigenvarieties and a p-ADic Labesse-Langlands transfer

Let B,G and G be as in Section 1 and let e := 0 ei G C'“(G(Ap,Q) be an
idempotent. Ghoose S D S{e) and dehne

nur,s := (g)'lRE(SL2(Qz),SL2(Z0),

l^S

where J^£;(SL2(Q/), SL2(Z/)) is the algebra under convolution of compactly sup¬
ported E-valued functions on SL2(Q/) that are bi-invariant under SL2(Z;). Denote
by I the Iwahori subgroup of SL2(Qp) given by

^ ) G SL2(Zp) : c = 0 mod p| .

Let

'Hs := 'Hur,S ®E •A.p

be the Hecke algebra, where Ap is the commutative E-subalgebra of the Iwahori
Hecke algebra J^£;(SL2(Qp), /) generated by the characteristic function on the double

4

JUDITH LUDWIG

coset

We let W := Homcts(Z*, Gm) be the usual weight space. Using the spaces of over-
convergent forms for G as constructed in [9] and Buzzard’s machine, one can attach
to this data an eigenvariety P(e,5) of idempotent type e (cf. [1], [9]).

To be precise, one also has to make a choice of a compact operator in the con¬
struction, which we fix once and for all to be

uo ■= ® li(p-i p)i ^ ^5-

The eigenvariety ^(e, S) is a rigid analytic space defined over E and it comes
equipped with a locally finite (on the source) morphism

w : P(e,5) ^ W

and an iii-algebra homomorphism

^l;:ns^O{V{e,S)).

The points of P(e, S) correspond to finite slope systems of Hecke eigenvalues occur¬
ring in the space of overconvergent forms mentioned above. Moreover the map

V{e,S){%) ^ Hom(i^s,Qp) x>V(Qp)

X ^ {'ijjx{h) :='il>{h){x),uj{x))
is an injection (cf. Lemma 7.2.7 of [T]).

Likewise, starting from an idempotent e E C“(G(A^),Q), a set 5 D S'(e) and an
associated Hecke algebra Us = 'Hur,s®E-^p, we can build an eigenvariety Vie, S) of
idempotent type e for G. The weight space in this case is W := Homcts((Zp^, Gm)-
We have natural maps (see cni Sections 2.2 and 2.3])

/r : VV ^ W,

A : T-Ls ^ Tis-

In the construction of Vie, S) we always choose A(no) as the compact operator.

There are natural Q-structures on the Hecke algebras defined by the subalgebras
of Q-valued functions which we denote by ^Q,nr,s etc. The monomorphism

A can in fact be defined over Q (see Section 2 of m for details). In particular we
constructed in Lemma 2.10 of m an inclusion of Q-algebras

Az,q : iI^Q(SL2(Q/),SL2(Z0) i«Q(GL2(Qz), GL2(Zz)).

We also refer to m for details regarding the construction as well as general
properties of the eigenvarieties.

Remark 2.1. There exists a Zariski-dense and accumulation subset

ZcV{e,S){%)

coming from p-refined classical automorphic representations as defined in [TO], Def.
3.14- This is proved as usual (cf. |5l Section 6.4.5] and also [TOl Prop. 3.9] for

L-INDISTINGUISHABILITY ON EIGENVARIETIES

5

a proof of the analogous assertion for G in the same notation) using the fact that
forms of small slope are classical and that classical weights are dense in weight space.

Define ti G 'Hur,s, as the characteristic function on the double coset

SL2(Z^) ( q ) SL2(Z^),

where ;-i) is understood to be the matrix in SL 2 (Z‘^) = W^^gSlj 2 { 1 :q) which is
equal to 1 for all g / Z and equal to ;-i) at 1. Furthermore let

hi := y + 1) G T-Lur,s-

Note that hi is an element of the subalgebra g C 'Hur,s of O^-valued functions.

Lemma 2.2. Let V{e, S) he an eigenvariety of idempotent type for G. Then there
exists a 3-dimensional pseudo-representation

T:GQ,s^O{V{e,S))
such that T(Frob;) = if{hi) for all I ^ S.

Proof By the previous remark we have a Zariski dense subset Z C P(e,5)(Qp) of
classical points. A point z £ Z comes from an algebraic automorphic representation
TT of G{A) and there is a projective Galois representation

Pz ■ Gq^s PGL 2 (Qp)

associated to tt. Namely, if vr is any algebraic automorphic representation of G{A)
which is unramified outside S and lifts tt, let p{tt) : Gq^s GL 2 (Qp) be the attached
Galois representation by Deligne. It has the property that for any I ^ S, the
characteristic polynomial of /9(7f)(Frob/) is given by — Ti(jt)X + ISi{tt), where
Ti{7r) := ip o and pi is the eigenvalue of

■“ ^gl2(2s)(* ^)gl2(2-s) ^

on (^pGL2(z5) and similarly for Si = ) ouizS) ^

Then

Pz = r]o p{7r) : Gq,s PGL2(Qp)

is the composition of p{tt) and the natural homomorphism rj : GL 2 (Qp) —)• PGL 2 (Qp).
There is a monomorphism i : PGL 2 (Qp) ^ GL 3 (Qp) coming from the adjoint action,
which identifies PGL 2 (Qp) with S 03 (Qp). Dehne

az ■— io Pz ■ Giq^s GL3(Qp).

Then

az = Sym^(p(7r)) (g) det{p{7r)) ^

and an easy calculation shows that for all Z ^ S

Tr(c7^(Frobi)) = Tr2(p(7r)(Frobz))/det(p(7r)(Frobz)) - 1
= T^in)/ilSiin))-l.

6

JUDITH LUDWIG

Now an elementary calculation shows that Tf /{ISi) — 1 = XQ{hi) and therefore

The lemma now follows from Proposition 7.1.1 of [5] (see also Section 3.1.3 of m),
i.e., Hypothesis H in [5] is satisfied using the Zariski-dense set Z C P(e, 5)(Qp), the
representations for z G Z and the family of functions £ Cl(P(e, S)). □

Remark 2.3. In the following we abbreviate T>{e) := 'D{e,S{e)) and 'Dfe) :=

Vie, 3(10).

Below we will often use so-called speeial idempotents attached to a finite set of
Bernstein components. Let F/Qi be a hnite extension, H(F) the T-points of a
reductive group over F. Given a Bernstein component s of the category of smooth
Q-representations of H(F), there is an idempotent G C^(H(F),Q) such that for
an irreducible smooth Q-representation a of H(F), • cr 7 ^ 0 if and only if a is

contained in the Bernstein component s. Similarly one can attach an idempotent
to a finite set S of Bernstein components. We refer to Section 3 of [3] for a nice
overview and to Proposition 3.13 of loc.cit. for the existence of these so-called special
idempotents.

When we choose special idempotents below, we may always assume they take val¬
ues in Q as all automorphic representations we deal with in this paper are algebraic
and have the property that their hnite part is dehned over Q.

We also want to remark here that irreducible supercuspidal representations that
are in the same Bernstein component differ from each other by a twist by an un-
ramihed character. In particular, if two irreducible supercuspidal representations a
and a' of SL 2 (Qz) are in the same Bernstein component, then cr = a'.

We use the following notation: Let s be the Bernstein component of G(Qi) dehned
by a supercuspidal representation tti of G(Qi). Then we denote by ResQ(s) the hnite
set of Bernstein components dehned by the representations occurring in
Note this is well dehned.

The classical transfer, by which we just mean the map that attaches to an auto¬
morphic representation tt of G(A) an L-packet of representations n(7f) of G(A), can
be interpolated to maps between suitable eigenvarieties.

Definition 2.4. Two idempotents e G C'^(G(Ap,Q) and e G C'^(G(Ap,Q) are
called Langlands compatible if they satisfy: For any discrete automorphic represen¬
tation TT ofG{A) with e-TT^ ^ 0 and any r G n( 7 rp), there exists an element tt in the
packet n(7f), such that

• m(7r) > 0,

• e • TTj / 0 and

• TTp = T.

Theorem 2.5. (1) Lete£ C'^(G(Ap,Q) be an idempotent. Then there exists

an idempotent e G C'“(G(Ap,Q) with S{e) = 5(e) and such that e and e
are Langlands compatible. Define S:=S(e).

L-INDISTINGUISHABILITY ON EIGENVARIETIES

7

(2) Assume e = ®ei E (^“(^(ApjQ) has the property that for all I E ^(e), ei

is a special idempotent attached to a supercuspidal Bernstein component S(.
Define e := ®ei E C'“(G(Ap,Q) where e/ := esL2(Z() ^ ^ ‘S'(e) and

/or Z G S{e}, ei is a special idempotent attached to the finite set of Bernstein
components Resg(s/). T/ien e and e are Langlands compatible.

(3) For any two Langlands compatible idempotents e E C'^{G{AiF^),Q) and e E
C'“(G(Ap,Q) with the same set S of bad places there exists a morphism
( : D(e) —)■ D(e) such that the diagrams

P(e) —^P(e)

ns

^ns

w

-+W

c

0{V{e))^0{V{fi))

commute.

Proof. Part (1) follows from Proposition 4.15 and Proposition 4.16 of [TO], once we
remark that for any idempotent e E C“(G(Ap,Q), there exists a compact open

subgroup K C G'(Aj) such that ■ e = e = e ■ ej^. Part (3) follows from [TO]
Theorem 5.7 and the proof of it.

Part (2) can be proved in the same way as Proposition 4.16 of [TUj. Namely for
an automorphic representation T of G(A) with e 0 and any r E n(7fp) define

Y{n,T) := {vr E n(T)|e • (vrp / 0,7rp = r}

= {tt E n(7r)|7ri = vrf VZ ^ S'(e),7rp = r},

where in the last line denotes the unique member of the L-packet 11 (tt^) with
( 7 r[’)^'" 2 (Zi) ^ Q_ need to show that there exists vr E Y{n,T) such that m(7r) > 0.
For that let vr E Y{n,T) be arbitrary and assume m(7r) = 0. Then by Proposition
4.11 of |10] we may change vr at a prime Z E Ss to a different representation in the
local L-packet n(T;) to get a representation tt' which is automorphic and still in
Y{^,t). □

Remark 2.6. ITe recall that / is constructed using two auxiliary eigenvarieties
D'{e) and V"{e), which are described in Section 3.3 and 3.f of [TO]. D'fe) is the
eigenvariety which apart from the Hecke algebra is build from the same data as D(e)
but where the Hecke algebra is replaced by %$■ A comes equipped with a morphism
w' : L>'(e) —)• W and the points ofV'fe) embed

Hom{ns,Qp) X W(Qp).

We have a morphism X' : D{e) —)■ D'(e), which on points is given by

JUDITH LUDWIG

The second eigenvariety V''{e) is simply defined as the pullback W Xw P(e) and
the morphism C, is the composite

V{fi) V"{e) -> P(e) .

W

id

id

By Proposition 5.6 of m, the morphism is a closed immersion. This is important
in what follows.

Remark 2.7. If e, e' E C-(G(Ap,Q) are two idempotents as above, such that
S{e) = S{e') and such that e'i\ei for all I E 5\{p}, then there exists a closed immer-
sionV{e) M-P(e'). (cf. [T] Section 7.3).

3. Slopes of CM points

We determine the slopes of points on eigenvarieties P(e) that arise from automor-
phic representations 7r(0) of G(A) coming from a Grofiencharacter.

We view as a subset of weight space via

Z2 W{Qp),{kuk2) ^ ({zuZ2) ^ z’l^zf

I : =

E GL2(Zp) : c = 0 mod p

Let k = (/ci, /C 2 ) E Z^, ki > k 2 . We denote by I the Iwahori subgroup of GL 2 (<Q;p
given by

a b
c d

Recall (cf. Section 7.2.2 of |1] and Definition 3.14 of [lOj l that a p-refined auto-
morphic representation of weight k of G{A) is a pair (tt, x) such that

• vr is an automorphic representation of G(A);

• TTp has a non-zero fixed vector under the Iwahori I and x = (xi)X 2 ) is
an ordered pair of characters Xi • Qp C*,i = 1,2 such that Tp

Ind^'"^^'®'’^(xi,X 2 ), where ) denotes the normalized parabolic

induction from the upper triangular Borel B C GL 2 (Qp);

• = (Sym^i-'=2(C2) ® Nrd^2)*_

Different points on an eigenvariety that come from the same automorphic representa¬
tion TT are parametrized by the different choices of pairs x = (xi; X 2 ), such that nf ^

Indg^^^'®'’^(xi,X 2 )j which are called refinements. Note if vr^ = Ind^'"^*''®^^(xi,X 2 ),
there are precisely two refinements, namely (xi)X 2 ) and (x2)Xi)-
Fix an idempotent e and let S := S(e). Define

The operator

'^ur^S

^1(1 p)I ^

‘■T(p-

is the image of uo under A, and we denote it by uq again.

L-INDISTINGUISHABILITY ON EIGENVARIETIES

9

Recall from Section 2, that R’(e) comes equipped with a morphism V’ : T~is
0(P(e)). For a point 2 ; on the eigenvariety P(e) which corresponds to a p-refined
automorphic representation (^, {xi,X 2 )) of weight k, we have

'fp{Up){z) = V’(5f,(xi,x2))(^>) =

and

( 1 ) 'iPMiz) = ip{x2ip)xi{p)~^)p''^~’'^^^ ■

To justify the next definition recall the following classicality theorem. Again
fix k as above and choose an affinoid neighbourhood X of in W. We denote
by M(e,k,k{X)) the Fl-Banach space of overconvergent forms of weight k and
‘tame level’ e as defined in Section 3 of [9]. It comes equipped with an action
of Tis- We sometimes omit the parameter k{X) and write M(e,k) instead. If
X G P(e)(Qp) with oj{x) = k, then there exists an overconvergent finite slope eigen-
form / G M{e,k,k{X)) with eigenvalues 'ipx- The space M(e,k,k{X)) has a finite
dimensional subspace M(e, of classical forms.

Theorem 3.1 (0 Theorem 3.9.6]). Let k = (fei, ^ 2 ) G 7?, ki > k 2 - Let E'/E be a
finite extension, A G E'* and a := Vp{X). If

a < ki - k 2 + 1,

then the generalized X-eigenspace of Up acting on M{e,k,k{X))®EE' is contained
in the subspace M{e,kY^^EE'.

Definition 3.2. (1) A point x = (■i/’x,w(x)) on T’(e), with uj{x) = {ki,k 2 ) is

called of critical slope if Vp{fix{Up)) = /ci — A :2 + 1 .

(2) A refinement X of an automorphic representations ofG{A) of weight {ki, k 2 )
is called of critical slope if Vp{'ip(^^y^'){Up)) = ki — k 2 + 1-

Now let L/Q be an imaginary quadratic extension and let 9 : A'^/L* —)• C*
be a Grofiencharacter which does not factor through the norm. In [7], Jacquet
and Langlands show how to associate to 0 a cuspidal automorphic representation
t { 9 ) of GL 2 (A). We refer to §12 of [7] for details regarding the construction and
characterization. Assume t { 9 ) is in the image of the global Jacquet-Langlands
transfer JL from G to GL 2 , i.e., r(0)^ is a discrete series representation for all v G Sb-
Then s{9) := JL“^(r(0)) is an automorphic representation of G'(A). Assume s{6)
is of weight {ki,k 2 ) G so

7r(0)oo = (Sym*’i-^2(c2) (g, Nrd'^^)* ^ Sym*’i-'=2(c2) ® Nrd"*’!

and 0OO : —)• C* is given by Oooiz) = (] 2 ;^)-fci-i/ 2 ^fci-fc 2 +i ("ggg Remark 7.7 of [ 6 ]).

Let e G C'^(G(A^),Q) be an idempotent such that e • '/r(0)^ 7 ^ 0.

Lemma 3.3. Let s{6) be an automorphic representation of G{A) associated to a
Grofiencharacter of L as above. Assume p splits in L and s{6)p is unramified.
Then s{9) has a refinement of critical slope. More precisely let x, y G Vifif) be the

10

JUDITH LUDWIG

two points attached to 7r(0). Then the slopes of and 'tpy{Up) are fci — A :2 + 1

and 0. Furthermore

Vp{ifx{uQ)) = 2{ki - k2 + 1 ).

Proof. The GroBencharacter 9q := 0|| is algebraic and we can turn it

into a p-adic character by shifting the weight from oo to p, i.e., we define

e' : Al/L* ^ Q;

efx) = ip{9o{x)0Ql^{Xoc))T^o{Xw)~^^Tw{Xw)~’^^~^,

where we have matched the two complex embeddings of Loo with the two places w,w
above p. The finite part of an algebraic Grofiencharacter takes values in a number
field. Moreover, 6' factors through the compact group A2/T*T^, so it takes values
in 0*p for some finite extension F/<Qp.

Our assumptions imply that vr(0)p = and the two refinements

of tt{9) are given by (6'^,6'w) and {9^,9^)-

Let pw = G A*p (respectively pw) denote the idele which is 1

at all places except for w (respectively W), where it equals p. Then

O'ipvj) = ip(0w{p)p^^^)p~^^ =

^ (Pw) = t'p{dwip)p ^ )p ^ ^ ^ ■

As 9'{pw) and 9'{pjp) are in Op this implies the claim on the slopes of the Up-
eigenvalues. Using Equation m, one then verifies the slope of ipxiuo)- C

Remark 3.4. The character 9' : Af/L* —)> Q* in the proof of the above lemma is
trivial on so it factors through the quotient Ap/L*= Gff. We may therefore
view 9' as a continuous character ofGi with values in Q*. In this notation the Galois
representation '■ Gq —>■ GL 2 (Qp) attached to 'k{ 9) is given by

P . ie )=^- 4y ')

as one easily checks by comparing traces of Hecke operators and Frobenius.

Lemma 3.5. Let Tr{9) he an automorphic representation of G{A) of tame level e,
weight {ki,k 2 ) and unramified at p which is associated to a Grofiencharacter 9 :
A*p/L* —>■ C* and assume that p is inert in L. Then vr(0) gives rise to two distinct
points x,y on T>{e). Their slopes agree and are equal to

Vpi'ifxiUp)) = Vp{i)y(JJp)) = {ki -k2 + l)/2.

Proof. Let v denote the unique place above p, and let w : Q* —>■ C* be the character
associated to the quadratic extension L^/Qp by local class field theory. By assump¬
tion 9y is unramified and therefore factors through the norm : L* —)• Q*.

L-INDISTINGUISHABILITY ON EIGENVARIETIES

11

Let (5 : Q* —)■ C* be a character such that 0^ = S o Then by construc¬
tion 7r{0)p = (5a;), in particular, the two refinements are distinct. The

C/p-eigenvalues on the two points x and y are given by

i^xiUp) =

i)y{Up) = ip{5{p)uj{p)p^/‘^)p-’^^ = ip{-S{p)p^/^)p~’"\

In particular, we see that they have the same slope. A similar calculation as in the
proof of the last lemma, with

implies that the slope is given by {ki — k 2 + l)/2. □

4. Existence of L-indistinguishable forms

Let g > 5 be a prime number such that —q = 1 mod 4 and let L := Q{y/—q) be
the associated imaginary quadratic extension. Choose a prime p which splits in L
and let B be the quaternion algebra over Q such that Sb = {^jCo}- Let G and G
be as above. _

Let Tr{9) be an automorphic representation of G(A) coming from a Grofiencharacter
9 : —)• C* of L. Assume that

(1) 'k{9)i is unramihed for all I / q.

(2) The L-packet n(7r(0)g) = {ri,r 2 } defined by 7r{0)q is of size two.

(3) Precisely one of the representations

VTl := vrf (g) n (g) TToo, 712 := vr° ®T2®'Koo
l^q l^q

is automorphic. As before ttJ^ denotes the unique member of the local L-
packet n(7r(0)/), which has a non-zero hxed vector under SL 2 (Zi).

Lemma 4.1. Automorphic representations 'k{ 9) ofG{A) satisfying the above list of
properties exist.

Proof. Note that (n«;^oo^i ^ ^qo)/^l ^ A*j^/L* is of finite index. Furthermore,
our assumptions imply that Of = {1,-1} and that q is the only prime that ramifies
in L. Dehne a Grofiencharacter 9 : Af/L* C* as follows:

Let 0OO '■ L*^ —)■ C* be the character given by 9oo{z) i-A (zzYz^, where r G C and
m > 2 is an even integer, so that 9oo is trivial on Of.

Denote by v the unique place of L above q. For all re / u let 9^ ■ Of —)■ C* be
the trivial character.

For V , let O^ be the kernel of the norm map (N2,„/q^)|c)j . Ghoose any continuous
non-quadratic character

9',:0^/{l,-l}^C*
and extend it to a continuous character

0^:01^ CL

12

JUDITH LUDWIG

Now the character

Il9^x9^: I II O*^ XL* \/Ol^C*

w^oo \w^oo j

is continuous. Extend it arbitrarily to a character 9 of L*\h^^.

We verify the conditions (l)-(3) for vr(0). By construction 0^, is unramified for
all finite places w not equal to v, and the local extensions L^/Qi are unramified for
I, which implies (1). Part ( 2 ) follows from Lemma 7.1 of [ 8 ]. As we have chosen a
non-quadratic character in the construction of 9^, there exists an element 7 G
such that 9jj(j) / 9jj('y~^) = 9y(j). The other conditions of Lemma 7.1 of [ 8 ] are
also satisfied by construction.

Part (3) follows from the multiplicity formulae. The representations vri and 712
are of type (a) and the formula for their multiplicity is given in Proposition 7.3
of [8]. □

Remark 4.2. The reason for this slightly delieate choice of the local character at
the place above q in the above proof is that we are eonstrueting L-paekets of an
inner form of SL 2 , whieh is not quasi-split. Changing a representation in a global
endoseopic packet at a plaee where the local L-packet is of size two therefore not
always changes the multiplicity.

Now fix an automorphic representation 7 r( 0 ) as above and such that

9^{z)

where fci, /c 2 G Z and /ci — A :2 + 1 > 2 is an even integer. In particular,

tt{9)oo = ^ Nrd^^)^

We have two representations vri and 712 as above and we assume that 712 is automor¬
phic and TTi is not.

The representation 7 r( 0 ) shows up in the following eigenvariety: for all 1 7 ^ g
define ei := eGL 2 (Z() let Cq be the special idempotent attached to the Bernstein
component defined by the supercuspidal representation 7r(0)q. Define e = G
C^{G{A^jr),Q). Let S = S(e) = {p,q} and LLs ■= TLur.s ® Ap as in Section 2. Then

by construction tt{9) gives rise to two points on the eigenvariety T>(e), one of which
is of critical slope by Lemma 13.31 which we denote again by x.

Let (respectively G C^{G{Qq),Q) be the special idempotent associated
with Ti (respectively T2) and define

ei := esL 2 (Zi) ® G C^{G{APj),Q) and

62 := (g) esuiz,) ® eg,2 G ^^(^(Ap,^).

i¥^q,p

Theorem 4.3. There exist points xi G D(ei)(Qp) and X2 G P(e 2 )(Qp) such that
{fj^^^ixixi)) = {'ifx^,u}{x 2 )) = (V’s|hs,/^(w(x))).

L-INDISTINGUISHABILITY ON EIGENVARIETIES

13

Proof. By construction tt2 is an automorphic representation such that 62 • 7 ^ 0

and so there is a point X 2 G 'D{e 2 ){Qp) as claimed.

For the existence of xi we use the p-adic transfer. Recall the notation of Re¬
mark ESI We have a Zariski-dense and accumulation set Z' on 'D'fe), which is in
bijection with the set of pairs

{(A o'0(5f) (fcl) ^ 2 ))})

where ^ is a p-rehned automorphic representation of G{A) of weight {ki,k2) (cf.
Section 3.3 of [ID]). In particular, X'{x) G Z'.

Let IIs be the set of all stable L-packets of G{A) and let

:= {(A' o (fei, fes)) G Z' I n(i) G n,}

be the subset of Z' arising from representations ^ that do not come from a Grofien-
character. This is well-de fin ed (cf. [TU] Section 3.3.1).

Claim: There exists an open affinoid neighbourhood U of A'(x) G T>'(e){Q_p) such
that Zgr\U is Zariski-dense and accumulation in U. Indeed choose any open affinoid
neighbourhood W of A'(x) and let

V := {x £ IT| Vp{f;x{uo)) = Vp{'ipy{x)iuo)) = 2 {ki -k2 + 1)}.

This is an affinoid neighbourhood of X'ifc). Choose 1/ C R to be an open affinoid
neighbourhood of X'ix) with the property that oj'{U) C W is open affinoid and the
induced morphism uj'\u : U —>■ oj'{U) is finite and surjective when restricted to any
irreducible component of U.

To see that Z' n 17 is Zariski-dense and accumulation, let y := fj,{ui(x)) = ki — k2
and y' := 2 {ki — ^2) -|- 1 G W{E) and define Y := {y,ouj')\fj^[{y,y'}) to be the fibre,
which is a Zariski-closed subspace of U of codimension 1. Let U' := U\Y. Then
Z'CiU' C Zg by Lemma ESI and Lemma 1531 above. But Z'CiU' is still Zariski-dense
and accumulation in U, as we have only removed a Zariski-closed subset of smaller
dimension. This proves the claim.

Define e := p 6 sL 2 (Zi) ® G C'^(G(Ap,Q), where Cq is the special idempo-
tent associated with the two Bernstein components defined by the representations ri
and T2. Then e and e are Langlands compatible and we have a p-adic transfer as in
Theorem 12.51

By Remark 12.71 we have a closed immersion 77(ei) 77(e), which we base-change
along ^ : W —)• W to t : P"(ei) V"{e). Consider the following diagram (cf.
Remark 12. 6 p

77"(ei)->77(ei) .

L

77(g) 77'(g) P"(e)-^ P(e)

As ^ and i are closed immersions of equi-dimensional rigid analytic spaces, their
images are a union of irreducible components of 77"(e). We identify 77'(e) and 77"(ei)
with their images in 77"(e), i.e., we consider them as subspaces of 77"(e).

14

JUDITH LUDWIG

Let T be an irreducible component of containing A^(5:). Let U be as above.
Then UCi Z^CiT is still Zariski-dense in [/n T and so there exists a point s G Z'^CiT
and we can also assume that s ^ T' CiT for any irreducible component T' ^ T.
As s G Z', s comes from an automorphic representation that gives rise to a stable
L-packet for G, which implies that s G P^^(ei). Therefore T C and in

particular, A'(x) G P"(ei). □

Remark 4.4. One can of course cook up other examples by using more general
imaginary quadratic fields L and quaternion algebras B with more ramified primes.
For example assume that B is ramified at more places and tt' is a representation
in an endoscopic L-packet n(7f), such that 7r[ is unramified for all I ^ Sb, with
m(7r') = 0, and such that n(7fp) has size one. Then one can again use the special
idempotents at the bad places to construct idempotents e G C^{G{A^j),Q) such that

e • TTj / 0 for IT G n(7f) if and only if ir = it'.

In fact this trick works as long as tt[ is supercuspidal at all places where it is not
unramified.

Corollary 4.5. In the notation of Theorem \4-S\ define (p := fixlus ^ ~

pi{uiix)) = ki — k 2 ^ W{E). The eigenspaces and M(e 2 ,n)‘^ are both

non-zero. The Galois representations attached to the eigenforms in these two spaces
agree.

Proof. The Galois representations exist by Lemma r2.2l and depend only on the points
on the eigenvariety defined by the eigenforms. But the images of xi and X 2 in T’(e),
with e as in the proof of Theorem 14.31 agree by construction. □

5. Consequences

Definition 5.1. Let to : V{e) —>• W (respectively u : T>{e) —>• W) he an eigenvariety
of idempotent type e (respectively e). We call a point z G T>(e)(Qp) (resp. V{e){Qp))
classical if there exists f G M{e,oj{z)Y^ (resp. M{e,dj{z)YY such that h- f = 'ifz{h)f
for all h G TLsie) (resp. 'Hs(e))-

For the group G we have the following phenomenon.

Proposition 5.2. Assume e' ande in GY{G{Af),Qp) are idempotents with S(e!) =
S'(e) =: S and assume for all I G S the local idempotents and ei are special
idempotents associated to Bernstein eomponents. Assume that e*e! = el = 'S *e so
that we have a closed immersion h : P(e') Vie). Assume z G P(e')(Qp) is such
that h{z) is classical. Then z is classieal.

Proof. We have to show that there exists a classical automorphic eigenform / G
M(e',uj{z)Y^ with system of Hecke eigenvalues fiz- By construction of the eigen-
varieties there exists an overconvergent eigenform foe G Mife!,u){z)) with system
of Hecke eigenvalues given by fiz- By assumption the point h{z) is classical, so
there exists an eigenform g G M(e,u}{z)Y^ for the same system of Hecke eigenvalues
Yhiz) = Yz- In particular, we have a p-rehned automorphic representation

L-INDISTINGUISHABILITY ON EIGENVARIETIES

15

giving rise to h{z). Both eigenvarieties T>{S) and P(e) carry pseudo-representations
T' and T (see Prop. 3.10 of [TOl) and T' = h* o T, where h* : 0(P(e)) ^ 0(P(e'))
denotes the homomorphism induced by h. For / G 5 let I; be the inertia subgroup of
Gal(Q//Q). By Lemma 7.8.18 of [T], T\J^ is constant on the connected components
of ^(e}. Let X G 'D(e’) be a classical point on the same connected component as 2 .
Then T^\ii = Local-global compatibility, compatibility of the local

Jacquet-Langlands transfer with twists and the inertial local Langlands correspon¬
dence (see Appendix 1.2 of m) imply that the local components {T^h{z))i and {'Kx)i
are in the same Bernstein component. Therefore • {^h{z))i 7^ 0- ^

The situation for eigenvarieties of the group G is different. First of all we have
the following result:

Proposition 5.3. In the notation of Theorem 4-3. the point xi G P(ei)(Qj;
classical.

is not

Proof. Assume xi is classical. Then there exists an automorphic representation vr
of G(A) snch that

(2) ei • TT^ / 0,

in particnlar ^ Q for all I ^ {p,q}- The system of Hecke eigenvalnes

'4’x\HurS(r) determines the representation vr; with ^ q local L-packet

11/ = n(7r(0)/) uniquely. But this implies tt G n(7r(0)) (cf. Theorem 4.1.2 of [TT]i.
Condition ([2]) implies that vr^ = ri, so the only choice left might be at p. But by con¬
struction n(7r(0)p) = {vTi^p} is a singleton and therefore vr = tti. But m(7ri) = 0. □

Remark 5.4. It is obvious that one cannot produce non-classical points starting
from Grofiencharacters of an imaginary quadratic field L in which p is inert. There
are multiple reasons for this. For example, note that any ‘candidate for xi ’ that
one would end up constructing would automatically be classical by the classicality
theorem. The critical slope for uq is given by 2{ki — /c 2 + 1) and Lemm.a, \3. ,71 implies
that the slope of the UQ-eigenvalue of any candidate is ki — k2 + l.

Corollary 5.5. Let V{e) be an eigenvariety of idempotent type e for G and assume
z G V(e){Qp) is a point whose system of Hecke eigenvalues comes from a classical
automorphic representation n of G{A). Then z is not necessarily classical.

Remark 5.6. Note however that one can always enforce classicality by passing to
a suitable idempotent (e.g., the idempotent attached to a sufficiently small compact
open subgroup K C G(A ^)). In our example, the image of the non-classical point xi
under the map T){ei) V{e) from the proof of Theorem \4.^ is classical. The
analogue of Provo.sition HOI for G is therefore false.

References

[1] Joel Bella'iche and Gaetan Chenevier. Families of Galois representations and Selmer groups.
Asterisque, 324:xii+314, 2009.

16

JUDITH LUDWIG

[2] Christophe Breuil and Ariane Mezard. Multiplicites modulaires et representations de GL 2 (Zp)
et de Gal(Qp/Qp) en / = p. Duke Math. J., II5(2):205-310, 2002. With an appendix by Guy
Henniart.

[3] Golin J. Bushnell and Philip C. Kutzko. Smooth representations of reductive p-adic groups:
structure theory via types. Proc. London Math. Soc. (3), 77(3):582-634, 1998.

[4] Kevin Buzzard. Eigenvarieties. In L-functions and Galois representations, volume 320 of Lon¬
don Math. Soc. Lecture Note Ser., pages 59-120. Gambridge Univ. Press, Gambridge, 2007.

[5] Gaetan Ghenevier. Families p-adiques de formes automorphes pour GL„. J. Reine Angew.
Math., 570:143-217, 2004.

[6] Stephen S. Gelbart. Automorphic forms on adele groups. Princeton University Press, Princeton,
N.J.; University of Tokyo Press, Tokyo, 1975. Annals of Mathematics Studies, No. 83.

[7] H. Jacquet and R. P. Langlands. Automorphic forms on GL(2). Lecture Notes in Mathematics,
Vol. 114. Springer-Verlag, Berlin-New York, 1970.

[8] J.-P. Labesse and R. P. Langlands. L-indistinguishability for SL(2). Canad. J. Math., 31(4):726-
785, 1979.

[9] David Loeffler. Overconvergent algebraic automorphic forms. Proc. Land. Math. Soc. (3),
102(2):193-228, 2011.

[10] Judith Ludwig. A p-adic Labesse-Langlands transfer. 2014. arXiv: 1412.4140.

[11] Dinakar Ramakrishnan. Modularity of the Rankin-Selberg L-series, and multiplicity one for
SL(2). Ann. of Math. (2), 152(1):45-111, 2000.

E-mail address: jludwig@math.uni-bonn.de